Inclusion and exclusion criteria

In addition to ensuring included studies were strictly responding to the research question, inclusion and exclusion criteria were also designed to exclude any study which demonstrated any risk of bias. These criteria are described in more detail in the corresponding sections below.

In conjunction with the study question, only studies aimed at adults were used. Adulthood was defined by both legal and employment norms – i.e., a minimum of 18 years to a maximum 65 years of age. The age of 65 was selected as the upper limit because it is the age of retirement in many developed countries, while the lower limit of 18 is typical for the majority of legal recognitions. Studies with a range including above 65 (e.g., 22 to 75 years old) were included so long as the median age of participants remains below 65. In contrast, any study including children under 18 as well as adults was not included, unless the data was detailed in sub-group analysis allowing the researchers to only use adult data. Children under the age of 18 are the subjects of many studies, but these interventions would likely not be applicable to the general population as they would predominantly take place in schools. Finally, studies were not required to cover this entire age range, but were allowed to cover only a portion (e.g., 18–22 year olds).

Experiments were also required to include both men and women for the sake of general population value of the results. Numerous studies have noted that men and women respond differently to dietary interventions [12–14], and results obtained from only one gender may therefore be skewed. Indeed, many of the papers reviewed at the full-text and abstract level showed different results and levels of significance for the men and women included in the study. Studies using convenience sampling methods were not required to report the sample’s gender balance. For laboratory-based trials, however, included studies were not permitted to have a ratio higher than 70–30 in either direction.

Studies were excluded if participants were exclusively patients of a specific medical ailment, such as diabetes. Similarly, the subject pool could not be limited to obese or overweight people, as they would have well-documented differences in appetite, satiety, and restraint ratings [15]. Laboratory studies were required to exclude already-dieting participants, as they may have abnormal baseline eating patterns. Several studies included restraint ratings in their participant characteristics. Results for “restrained” and “unrestrained” eaters were pooled in some cases to remove this confounding effect.

In summary, included studies had to be aimed at a generic population, and not be primarily focused on a subset (e.g., women, diabetics, current dieters, etc.). Studies which reported sub-group analyses were not automatically excluded, provided they reported aggregated results or sufficient raw data to independently combine sub-group results in secondary analysis.

Studies must have been aimed at influencing behaviour relating to food consumption. Those focusing exclusively on beverages or alcohol consumption were not accepted. All study designs were accepted given they included a form of control or baseline comparator. For studies with parallel intervention groups, allocation of interventions had to be randomised, and baseline characteristics reported to ensure no significant pre-treatment group differences. Qualitative studies were excluded, though papers providing both qualitative and quantitative data were included so long as the quantitative data was reported according to the other standards described in this section.

Fundamentally, the intervention of interest must change the choice architecture while maintaining the autonomy of the test subject. Alterations to choice architecture include changes to the environment (e.g., olfactory or social), perception (e.g., emotional priming), availability of food (such as convenience or portion size), or knowledge-based changed (e.g., labelling). Any intervention which involved directly asking tests subjects about the experimental condition was excluded, as it would bias the participants. Similarly, interventions were not accepted which changed the fundamental properties of the food, such as energy density or fat content. Food appearance could be altered, however, such as a change in food unit size, colour, or odour. No options could be forbidden to the consumer or test subject, and no direct financial incentive to a participant could be involved.

Many of the reviewed and included studies examined several interventions simultaneously. For those examining both nudge and a non-nudge strategy, such as menu arrangement and pricing, respectively, results were required to be reported as a main effect of the nudge. If only the interaction effect was reported, the study was excluded.

Results had to be presented in terms of one of the following: calories, joules, grams, or purchases (either quantity purchased or a monetary amount). These had to be measured either directly or via purchase receipts; Participants were not permitted to report their own intake. Both absolute measurements and those reporting a change relative to baseline were accepted.

Studies which did not report a measurement of error – standard deviation, standard error, or confidence interval – were excluded. Studies that provided sufficient raw data for the independent secondary calculation of error, however, were included.